SiMa.ai raises $150M at a $1.45bn valuation with Fidelity and Michigan backing edge silicon

The proceeds go to two things: scaling Palette Neat, the company's software environment for what it calls Physical AI, and building next-generation silicon capable of 1,000 dense TOPS of compute, targeted for the first half of 2028 ^[9][10]. The hardware is two and a half years out. The round funds a roadmap the company has not shipped yet.

The revenue line is the part that moved institutional money. SiMa.ai said its revenue quadrupled from 2024 to 2025 and expects that to carry into 2026 ^[11]. It did not disclose the dollar figure, so the fourfold jump measures growth only; how big the business is remains unknown.

The investor list is what sets this round apart. Fidelity Management & Research Company co-led with Amplify Partners, and the new names include AllianceBernstein, Baron Capital, J.P. Morgan and the State of Michigan ^[3][4]. A pension-type balance sheet and two large asset managers are underwriting a chip company whose pitch is that it undercuts the incumbent. SiMa.ai said Physical AI applications have traditionally relied on Nvidia GPUs, "which can be expensive and power intensive," and that its full-stack platform simplifies deployments ^[14][15].

The company points to a $50 trillion market across humanoids, automotive and drones ^[16], and cites research projecting 145 million cumulative Physical AI device shipments by 2035 ^[17]. Both are the company's own framing. The harder number is the customer base: SiMa.ai said it serves more than 150 customers, naming Bosch, Emerson, Synopsys and others ^[12][13]. A hundred and fifty customers against a $1.45bn post-money is roughly $9.7m of valuation per customer. That ratio can't really be tested until the 2028 silicon ships and the revenue base is disclosed.

In the best case, edge inference becomes a real second market and the early institutional check looks cheap. Or Nvidia's power-and-cost disadvantage at the edge narrows before 2028 and SiMa.ai's differentiation erodes on the exact axis it sells on. Or the round is a growth-rate bet that stalls when the base is finally named. I lean toward the first being at least partly right, because the customer list is industrial and named. What would prove it wrong is simple: a shipped 1,000-TOPS part that misses its H1 2028 window, or a 2026 revenue figure that shows the quadrupling came off a base small enough to make the multiple absurd.

Total capital raised now sits at $500m, from a $30m Series A in 2020 ^[5][6][7]. The founder, Krishna Rangasayee, started the company in 2018 ^[8].